本研究提出一套以影像處理為基礎的交通速限標誌偵測與速限數字辨識系統,這套系統可自動提醒駕駛人目前的行車速度限制,以避免違規超速並確保安全。 本系統以顏色特徵為主、形狀特徵為輔,做速限標誌的偵測。我們以HSV色彩系統來偵測外框為紅色的速限標誌。為了降低天候狀況對偵測結果的影響,運用自適性閥值運算來解決不同光線的強弱所造成的影像問題,將偵測環境自動區分成白天和夜間兩大類,對於夜間的影像進行影像強化處理。接著以霍夫轉換偵測圓形特徵,若圓形大小在所設定的範圍內,則將其視為速限標誌。此外,本論文利用CMYK色彩系統做特徵值提取和影像強化,並進行形態學(Morphology)等影像處理,可以有效的切割出標誌中數字部份的影像,最後再以重新訓練後的Tesseract-OCR系統來辨識數字以獲取其速限資訊。 本研究以市售的行車紀錄器錄製晴天、陰天、夜晚、雨天等多種天候下的視訊,再以這些視訊測試所提方法的效能。實驗結果顯示,在偵測與辨識速限標誌上,於無下雨的白天狀況下,大約有92% 的準確率,而在夜晚也有80%的準確率。最後在一般的個人電腦下,每次含有交通速限標誌影像的偵測及辨識所花費的時間只需要0.2~0.4秒就能產生速限資訊。
In this study, a speed sign detection and digit recognition system is proposed based on image processing techniques. The proposed system can remind the driver of speed limits automatically to avoid traffic violation and keep driving safe. The system mainly utilizes the information of color features, along with shape features, to detect speed signs. First, we use the HSV color space to detect the speed sign with a red boarder. To avoid the influence of different weather or light conditions, we apply an adaptive threshold operation to deal with the images taken from various light conditions such that the captured images are automatically divided into two cases: day and night. When the detection event occurs at night, the proposed method will enhance the image of the traffic signs for a better result, followed by the detection of circle features with Hough transform. If the circle size is in the setting range, it is considered as a speed sign. Furthermore, image enhancement and feature extraction in the CMYK color space together with morphology operations are utilized to effectively cut out the digit part of a speed sign. Then the characters in the speed sign will be recognized by the retrained OCR software called Tesseract-OCR. We recoded the videos while driving by a commercial event data recorder (EDR) to evaluate the performance of the proposed method. The recoded videos contain several weather conditions such as sunny, cloudy, dusky, and rainy days. The experimental results show that, in the conditions of no rain and day time, the detection and recognition rate for speed signs is 92%, while it is 80% at night. Finally, the proposed system runs on a personal computer to perform the tasks of detection and recognition for the images containing speed signs, and it takes only 0.2~0.4 seconds to generate the information of speed limit.